Spark1.1推出了Uer Define Function功能，用户可以在Spark SQL 里自定义实际需要的UDF来处理数据。
因为目前Spark SQL本身支持的函数有限，一些常用的函数都没有，比如len, concat...etc 但是使用UDF来自己实现根据业务需要的功能是非常方便的。
Spark SQL UDF其实是一个Scala函数，被catalyst封装成一个Expression结点，最后通过eval方法计根据当前Row计算UDF的结果，源码分析见：Spark SQL源码分析之UDF
Spark SQL UDF使用起来非常方便，分2个步骤：
一、注册
当我们导入了SQLContext或者HiveContext，即有注册UDF的功能。
registerFunction(udfName : String, func : FunctionN)
由于scala语言的限制，这里UDF的参数仅支持22个。
二、使用
select udfName(param1, param2....) from tableName
三、示例
我们这里创建2张表：
第一张dual会从README.md读取记录，里面仅有一个字段line : String
第二张表src，有2个字段key,value，数据是spark sql自带的测试数据。
我们使用 sbt/sbt hive/console进入测试环境：
1、字符串取长度 len()
创建table dual:
scala> sql("create table dual(line string)").collect()
14/09/19 17:41:34 INFO metastore.HiveMetaStore: 0: create_table: Table(tableName:dual, dbName:default, owner:root, createTime:1411119694, lastAccessTime:0, retention:0, sd:StorageDescriptor(cols:[FieldSchema(name:line, type:string, comment:null)], location:null, inputFormat:org.apache.hadoop.mapred.TextInputFormat, outputFormat:org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat, compressed:false, numBuckets:-1, serdeInfo:SerDeInfo(name:null, serializationLib:org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe, parameters:{serialization.format=1}), bucketCols:[], sortCols:[], parameters:{}, skewedInfo:SkewedInfo(skewedColNames:[], skewedColValues:[], skewedColValueLocationMaps:{}), storedAsSubDirectories:false), partitionKeys:[], parameters:{}, viewOriginalText:null, viewExpandedText:null, tableType:MANAGED_TABLE, privileges:PrincipalPrivilegeSet(userPrivileges:null, groupPrivileges:null, rolePrivileges:null))
14/09/19 17:41:34 INFO HiveMetaStore.audit: ugi=root ip=unknown-ip-addr cmd=create_table: Table(tableName:dual, dbName:default, owner:root, createTime:1411119694, lastAccessTime:0, retention:0, sd:StorageDescriptor(cols:[FieldSchema(name:line, type:string, comment:null)], location:null, inputFormat:org.apache.hadoop.mapred.TextInputFormat, outputFormat:org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat, compressed:false, numBuckets:-1, serdeInfo:SerDeInfo(name:null, serializationLib:org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe, parameters:{serialization.format=1}), bucketCols:[], sortCols:[], parameters:{}, skewedInfo:SkewedInfo(skewedColNames:[], skewedColValues:[], skewedColValueLocationMaps:{}), storedAsSubDirectories:false), partitionKeys:[], parameters:{}, viewOriginalText:null, viewExpandedText:null, tableType:MANAGED_TABLE, privileges:PrincipalPrivilegeSet(userPrivileges:null, groupPrivileges:null, rolePrivileges:null))
载入README.md数据:
sql("load data local inpath 'README.md' into table dual ").collect()
scala> sql("select * from dual").collect()
res4: Array[org.apache.spark.sql.Row] = Array([# Apache Spark], [], [Spark is a fast and general cluster computing system for Big Data. It provides], [high-level APIs in Scala, Java, and Python, and an optimized engine that], [supports general computation graphs for data analysis. It also supports a], [rich set of higher-level tools including Spark SQL for SQL and structured], [data processing, MLLib for machine learning, GraphX for graph processing,], [and Spark Streaming.], [], [<http://spark.apache.org/>], [], [], [## Online Documentation], [], [You can find the latest Spark documentation, including a programming], [guide, on the project webpage at <http://spark.apache.org/documentation.html>.], [This README file only contains basic setup instructions.], [], [## Building Spark], [], ...
编写len函数并，注册函数：
scala> registerFunction("len",(x:String)=>x.length)
测试：
scala> sql("select len(line) from dual").collect()
14/09/19 17:45:07 INFO spark.SparkContext: Job finished: collect at SparkPlan.scala:85, took 0.072239295 s
res6: Array[org.apache.spark.sql.Row] = Array([14], [0], [78], [72], [73], [73], [73], [20], [0], [26], [0], [0], [23], [0], [68], [78], [56], [0], [17], [0], [75], [0], [22], [0], [67], [0], [26], [0], [64], [0], [21], [0], [52], [0], [44], [0], [27], [0], [66], [0], [17], [4], [61], [0], [43], [0], [19], [0], [74], [74], [0], [29], [0], [32], [0], [75], [63], [67], [74], [72], [22], [0], [54], [0], [69], [0], [16], [0], [84], [17], [0], [19], [0], [31], [0], [77], [76], [77], [77], [0], [67], [27], [0], [25], [45], [0], [42], [58], [0], [91], [29], [0], [31], [58], [0], [42], [61], [0], [35], [52], [0], [77], [79], [74], [22], [0], [51], [0], [90], [0], [16], [42], [44], [30], [17], [0], [0], [56], [0], [46], [86], [78], [0], [30], [0], [16], [0], [97], [70], [0], [0], [24], [0], [78]...
2、字符串连接concat_str
这里为了简单起见，就根据src表的key value类型 Int, String来做例子：
scala> sql("desc src").collect()
res8: Array[org.apache.spark.sql.Row] = Array([key,int,null], [value,string,null])
scala> sql("select * from src limit 10").collect()
res7: Array[org.apache.spark.sql.Row] = Array([238,val_238], [86,val_86], [311,val_311], [27,val_27], [165,val_165], [409,val_409], [255,val_255], [278,val_278], [98,val_98], [484,val_484])
编写并注册concat_str函数：
scala> registerFunction("concat_str",(a:Int, b:String)=>a.toString+b)
测试concat函数
scala> sql("select concat_str(key,value) from src ").collect()
14/09/19 18:17:22 INFO spark.SparkContext: Job finished: collect at SparkPlan.scala:85, took 0.082076377 s res28: Array[org.apache.spark.sql.Row] = Array([238val_238], [86val_86], [311val_311], [27val_27], [165val_165], [409val_409], [255val_255], [278val_278], [98val_98], [484val_484], [265val_265], [193val_193], [401val_401], [150val_150], [273val_273], [224val_224], [369val_369], [66val_66], [128val_128], [213val_213], [146val_146], [406val_406], [429val_429], [374val_374], [152val_152], [469val_469], [145val_145], [495val_495], [37val_37], [327val_327], [281val_281], [277val_277], [209val_209], [15val_15], [82val_82], [403val_403], [166val_166], [417val_417], [430val_430], [252val_252], [292val_292], [219val_219], [287val_287], [153val_153], [193val_193], [338val_338], [446val_446], [459val_459], [394val_394], [237val_237], [482val_482], [174val_174], [413val_413], [494val_494], [207val_... scala>
——EOF——
原创文章，转载请注明出自：http://blog.csdn.net/oopsoom/article/details/39401391
时间： 2025-01-08 01:10:08